DeepSeek, a China-based AI company has launched DeepSeek-R1-Lite-Preview for better reasoning and problem-solving capabilities. Announced in a post on X, the system is positioned as a competitor to industry leaders like OpenAI.
Commentators believe that DeepSeek’s transparency is ironic when companies in the West have not addressed these gaps.
DeepSeek AI’s new launch can do better math
DeepSeek, an AI company based in China, introduced a new version of its AI system called DeepSeek-R1-Lite-Preview. In a post on X, it said that the new AI system has improved reasoning and problem-solving abilities.
According to DeepSeek, the preview performs well on benchmarks like AIME (American Invitational Mathematics Examination) and MATH, which measure problem-solving and reasoning abilities. As the AI seems skilled at handling complex mathematical and logical problems, it might be ready to compete with OpenAI’s ChatGPT and specifically with OpenAI o1.
This week, Mistral AI’s Le Chat also announced updates to allow free access to new features. The launches are seemingly creating intense competition in the generative AI market, overcoming the shortcomings of the other.
DeepSeek thinks out loud unlike ChatGPT
DeepSeek says that its AI can show step-by-step real-time reasoning to make its thought process more transparent. Meanwhile, the AI company has also said that they will release the open-source model and API developer tools in the coming days.
According to a comparison chart cited by AI and tech commentator Andrew Curran, DeepSeek-R1-Lite-Preview achieves the highest score in parameters like AIME (52.5) and Codeforces (1450), outperforming competitors like OpenAI o1-preview and Claude 3.5 Sonnet.
It also leads in MATH-500 (91.6), indicating high performance in advanced problem-solving tasks. However, it lags in GPQA Diamond (58.5) and Zebra Logic (56.6) compared to models like OpenAI o1-preview (73.3 and 71.4, respectively). The figures mean that there is room for improvement in general knowledge and logical reasoning segments.
Two months after the o1-preview announcement, and its Chain-of-Thought reasoning has been replicated. The Whale can now reason. DeepSeek says that the official version of DeepSeek-R1 will be completely open source. https://t.co/Ya9mVyLvDP pic.twitter.com/6wZ8xoAyyz
— Andrew Curran (@AndrewCurran_) November 20, 2024
Cryptopolitan tried the features of the launch for an unbiased review. Firstly, DeepSeek’s chat requires a user login. The chat under the ‘Deep Think’ feature limits conversations to up to 50 messages per day. We can say that Deepseek thinks loudly while also estimating its time of response. It also solved the math problem we presented in a logical order. In comparison, ChatGPT 4o took less time for the solution but did not present a step-by-step reasoning for the same.
Influencer Bilawal Sidhu took a jibe at o1 and stated, “Ironic that OpenAI’s o1 model hides its chain-of-thought reasoning, while the Chinese DeepSeek-R1 makes it transparent to users. Shouldn’t it be the other way around?”
That said, China has a comprehensive framework around AI. On July 13, 2023, multiple Chinese authorities, including the Cyberspace Administration of China (CAC) and the Ministry of Education, introduced new regulations for generative AI technologies. These rules, called the Generative AI Regulation, officially came into effect last year on August 15.
The scope of the regulation reportedly covers the use of algorithms, deep synthesis technologies, the use of all generative AI technologies, and several other tech activities. And with its transparent reasoning approach, strong performance on competitive benchmarks, and plans to release open-source tools, DeepSeek is pushing the boundaries of generative AI in China and among its competitors globally.
Land a High-Paying Web3 Job in 90 Days: The Ultimate Roadmap